Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Would the 750m aid in heavy computational research?

J

Jodles

macrumors regular
Original poster
Dec 5, 2008
172
3
Hi all,

I'm wondering whether there is any point in getting the high end MBP for computational research like MATLAB, MaxMSP, Mathematica, Python, etc.?

If I understand correctly, the software would have to either be optimized for OpenCL (Iris Pro) or CUDA (Nvidia); and I believe MATLAB can take advantage of CUDA. But is there a significant performance jump to take advantage of here? Does anyone have particular experience with e.g. heavy MATLAB use with CUDA?

Thanks!

J
 
Consultant said:
Why not ask the software makers MATLAB, etc.
Click to expand...

I'm asking whether anyone has experience with graphics processing in computational stuff, on the MBP, comparing the Iris Pro to the 750m, OpenCL and CUDA. Sorry if that wasn't clear.
 
I doubt you will get an answer here. I do not use MATLAB, so I can't help you either, but I can say that Kepler graphics (like 750M) are known for fairly low compute performance. If the software you use supports OpenCL, the Iris Pro will be most likely faster.
 
Have you also considered your RAM requirements? I work with Matlab a lot, with Medical Images in particular and my RAM requirements as well as processing time often far exceed what the Macbook Pro can provide which is why I mostly do processing on dedicated nodes.

If you need to use CUDA then you will need the 750m no? Whereas if you are going the openCL route then you won't.
 
Jodles said:
leman: Thanks for your reply!

I came across this which would be informative to anyone else in my position: http://blog.accelereyes.com/blog/2012/02/17/opencl_vs_cuda_webinar_recap/

It doesn't fully answer the question, but it is a very interesting read.

J
Click to expand...

Yep, I know those slides. Basically, nowadays it matters less and less whether you use OpenCL or CUDA. Both APIs are virtually feature-equivalent, so in most cases, any performance difference comes down to the compiler performance. And modern compilers should be mature enough to perform good optimisation. There is a clear tendency towards more and more applications using OpenCL.
 
abta1 said:
Have you also considered your RAM requirements? I work with Matlab a lot, with Medical Images in particular and my RAM requirements as well as processing time often far exceed what the Macbook Pro can provide which is why I mostly do processing on dedicated nodes.

If you need to use CUDA then you will need the 750m no? Whereas if you are going the openCL route then you won't.
Click to expand...

Yes, my work is in audio (blind source separation, music information retrieval), so I would assume there's much less RAM requirements than for Medical Images (16GB should be more than sufficient, I definitely don't need to use dedicated nodes.. Or if I do I'll sneak my way into the astrophysics department...:D).

Whether I go CUDA or the openCL route is up to me.. I'm not locked in either way.. But it seems that openCL with the Iris Pro my be more than good enough for my purposes. Unless CUDA with the 750m would perform dramatically better. It's hard to compare without exact benchmarks on these two GPUs specifically...

leman: Based on your post it seems I'd be fine with going the openCL route and the Iris Pro. Maybe rather use the money saved on some fast backup...

Thanks guys!
 
devilcm3 said:
I assume that research softwares does not have OSX version of it? ( or they do ? )

If you want to maximize the performance, a gaming windows laptop will always be a better choice.
Click to expand...

That's true about performance. However, so much of the audio software we use is OS X only, and literally the entire research department is using OS X. Which is good because I use Windows 7 at work and I much prefer OS X...!
 
If your software uses CUDA for compute, then the 750M would be your only option.

If your software uses OpenCL for compute, then the Iris Pro can actually be faster based on the benchmarks.
 
Freyqq said:
If your software uses CUDA for compute, then the 750M would be your only option.

If your software uses OpenCL for compute, then the Iris Pro can actually be faster based on the benchmarks.
Click to expand...

Yep. See the benchmarks on the Bare Feats site.
 
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back